These are the features reconstruction and bookkeeping users ask for, or that competitors already advertise, each classified with the same honesty as your own list. Demand doesn't mean you should build it. Most of these are already done by someone, or are a trap for a solo founder.
Legend: TS = Table stakes, D = Differentiator, X = Distraction, L = Liability. Priority: P0 = v1 (first 90 days), P1 = months 4–9, P2 = months 10–24, — = don't build.
Sources: the demand signals come from competitor feature pages (DocuClipper, Valid8, Vyapar TaxOne, the 2026 extraction-tool comparison), the comparable's blog, and analyst judgment. No user interviews have been done yet, so every priority here must be re-checked against the day-1–14 calls in 08.
4.1 Documents and layouts: which to support first
| Document type | Who needs it | Category | Priority | Notes |
|---|
| US bank statements, top 50 layouts (big banks + largest credit unions) | Everyone | TS | P0 | Deterministic parsers. These cover most of the pages you'll see. |
| US bank and credit-union long tail | Tax resolution, divorce, forensic | D | P0 (LLM fallback) → P1 (saved layouts) | Where general tools fail. Every fixed failure becomes a saved layout. |
| Credit card statements | Everyone | TS | P0 | Same engine. Watch for multi-cardholder sections. |
| Scanned and photographed statements (OCR) | Forensic (discovery productions), estates | TS / D | P0 | Discovery productions are often scans. Balance proof matters most here. |
| Brokerage statements (holdings, activity, dividends, realized gains) | Estates, divorce, tax | D | P1 | Top 10 US brokerages. Separate sheets per section, with totals checked. |
| Loan and mortgage statements | Divorce, estates | TS | P2 | Simple once the engine exists. |
| Payslips | Divorce income schedules | X for now | P2 | Only on paid demand. |
| Invoices and receipts | Bookkeepers | X | — | Dext, Hubdoc and QuickBooks own it. |
| Non-English statements | Pivot B only | D (if tested) | — unless Pivot B | QuickBooks' native import is English-only. Check local competitors first. |
4.2 Demanded features, classified
A. Accuracy and proof (your core, so lean in)
| Feature | Demand signal | Category | Priority | Notes |
|---|
| Per-statement balance proof (opening + rows = closing; running-balance check per row) | High. "99% accurate and still useless" if a row is missed (2026 review). | D | P0 | The headline feature. Shown as a badge per statement and a column per row. |
| Row-to-page citation (file, page and position, clickable) | High among forensic and litigation users. Valid8 sells "verify against source evidence". | D | P0 | Cheap to build if designed in from day one. Painful to add later. |
| Exceptions queue (only failed statements need a human) | High | D | P0 | Turns "check 300 statements" into "check the 4 that failed". |
| Missing-month and gap detection per account | High in reconstruction | D | P0 | Statement date ranges must chain: this closing balance is the next opening. |
| Duplicate statement detection | Medium-high (discovery productions repeat documents) | D | P0 | Hash plus period and account match. |
| Side-by-side review (PDF page next to extracted rows) | High | TS | P1 | Expected by anyone who has used DocuClipper. |
| "99.9% accuracy" marketing claim | Competitors make it | L | — | Unprovable per customer. Show the proof instead. |
B. Case and bulk workflow
| Feature | Category | Priority | Notes |
|---|
| Case folders: bulk upload (ZIP, folder, 500+ files), auto-sort by account and period | D | P0 | The unit of work is a case, not a file. |
| Cross-account transfer matching (money from checking to savings shows on both sides) | D | P1 | Needed for deposit analysis, so internal transfers aren't counted as income. Firm plan. |
| Payee and counterparty normalization ("AMZN MKTP US*2K4..." becomes "Amazon") | TS / D | P1 | Helps summaries. Keep it simple and editable. |
| Deposits summary (by source, by month; transfers excluded) | D | P1 | The core of a bank deposits analysis for tax resolution. A summary, not advice. |
| Password-protected PDF handling | TS | P0 | Common with statements. Never store the passwords. |
| Email-in or shared-folder intake | TS | P2 | Convenience. Firms like it. |
| Fund tracing graphs, lifestyle analysis, fraud scores | X / L | — | Valid8 (enterprise) and DocuClipper Business already market these. They're the expert's job. |
| Expert-report drafting with AI | L | — | Liability if wrong, and it's their professional work product. |
C. Outputs and integrations
| Feature | Category | Priority | Notes |
|---|
| Excel workbook: one sheet per account, a summary sheet, a proof column, a citation column | D | P0 | The product people see. Get the format from the design partner. |
| CSV per account | TS | P0 | |
| QuickBooks (QBO) and Xero CSV import files | TS | P1 | For the catch-up buyer. QuickBooks now reads English PDFs itself, so this is convenience, not a moat. |
| IIF, OFX, QFX, QIF, QBJ | TS | P2 | Only on paid demand. |
| Tally XML, Zoho Books, Sage | X | — | Other markets. |
| API (for firms with their own tools) | TS / D | P1 (Firm plan) | The comparable's first big jump came from one customer who wanted a JSON API for bulk conversion. |
| Google Sheets export | TS | P2 | |
D. Security, privacy and trust (required for the buyer)
| Feature | Category | Priority | Notes |
|---|
| Auto-delete after 30 days (configurable), plus one-click delete of a whole case | TS | P0 | Retention is liability. |
| No training on customer files. Zero-retention model API terms. | TS | P0 | Publish it on the security page. |
| MFA, audit log (who uploaded, viewed and exported what, and when) | TS / D | P1 | Chain-of-custody comfort for litigation users. |
| Written security summary (encryption, sub-processors, incident response) | TS | P0 | Firms will ask. A one-pager is enough at first. |
| SOC 2 Type II | TS at the top of the market (DocuClipper has it) | P2 (only if Firm deals stall without it) | $15–40k+ and months of work. Not before $10k MRR. |
| Data residency options | X | — | Only relevant outside the US. |
E. Firm and team
| Feature | Category | Priority | Notes |
|---|
| Seats and roles (Firm plan: 5 seats) | TS / D | P1 | The ARPU lever: Solo $49 → Firm $149. |
| Client and case labels, billing-code export (pages used per case) | D | P1 | Firms re-bill the cost to the client. Show them what to bill. |
| Case passes (one-off, 60 days) | D | P0 | Stops occasional users from churning monthly, and captures their cash. |
F. Growth features (built into the product)
| Feature | Category | Priority | Notes |
|---|
| Free 3-page preview with a visible balance check | D | P0 | Shows the proof before payment, which is the whole pitch. |
| Public "does this statement balance?" checker (free, no signup, file deleted immediately) | D | P1 | A search magnet and a trust builder. |
| Bank-specific and use-case search pages with redacted sample outputs | D | P0 onward | The comparable's main channel. See 08. |
| "Report a bad row" button (the file goes to the test corpus, with consent) | D | P0 | Turns support into accuracy and layout memory. |
| Referral credit (free pages for each referred paying firm) | TS | P2 | Professional communities refer. |
| Affiliate program for bookkeeping influencers | X | — | They send the generic "PDF to Excel" crowd, who churn. |
G. The tempting ones (people ask; don't build)
| Feature | Why not |
|---|
| A mobile app | Reconstruction is desk work with big files. |
| Chat with your statements | A demo toy. Reconstruction users need tables and proof, not answers. |
| Categorization into a chart of accounts with AI | Bookkeeping platforms already do it after import. It's a liability in litigation work. Maybe P2 for catch-up users only. |
| Bank feed connections | That's the job you're deliberately not doing. Plaid costs money, and feeds are the platforms' territory. |
4.3 Net effect
Of roughly 45 candidate features, about 12 are real differentiators for reconstruction work, and about 10 belong in v1. Nearly all of them are about proof, provenance and bulk. None are about "better extraction". Extraction is the part you rent from a model and wrap in checks.